Design Agent · JLL

Design Decision Files

A prototype shows what the design is. It cannot show why, or what we tried and dropped. On my Corrigo prototypes the reasons lived in chat threads, meeting recordings, and my head, so every "how should this behave?" came back to me, from engineers, from new reviewers, and from every new session with an AI assistant. A teammate or the assistant would propose again what review had already decided against. Designers spend that time maintaining the design instead of on research and new solutions.

Company
JLL
Role
Design Lead
Period
2026

Decisions

01

The reasons live in the repo.

Each prototype keeps a docs folder next to the code: one index to read first, one file per area of the design, and a separate place for raw inputs such as research and meeting notes. The CLAUDE.md file at the root points the agent there, so it reads the answer before it answers.

02

One file per area, named by what a person sees.

A file called request-card stays current, where a file named after a meeting or a date goes stale. Each one says what was decided, why, and where it lives in the code, and ends with a status: decided or open, the date, and who reviewed it.

03

Every input records who asked and why.

The reason lets us weigh the ask later, or find a better answer to the same need. When nobody gave one, the file says "reason not given" and we go ask. An input becomes a decision only when it lands in an area file with a status.

04

Reversals stay on record.

When a decision is undone, the old line stays, marked superseded with the date and where the new one is. Without it, the next person adds the removed button back.

05

The agent writes, and the designer decides.

I paste in feedback, a meeting note, or a transcript, and the agent logs it, says which file it changes and what it reverses, and stops until I say go. The code and its decision file change in the same pull request, because a stale file is worse than none.

06

Each fact has one home.

The repo owns how the design behaves. Confluence keeps sign-off, Jira keeps the tickets, Figma keeps explorations, and each links to the repo instead of copying it, because a copy kept by hand drifts.

Outcome

I set this up on my own Corrigo prototypes first, Service Agent and the Corrigo Enterprise web prototype, then wrote it up as a one-page guide with prompts a designer can run on their own repo, in order. On those prototypes an engineer, a reviewer, or the agent can answer how something should behave, and why, from the repo without asking me. I am bringing it to the rest of the design team, so the maintenance runs on its own with a designer deciding, and designers get their time back for research and new solutions.

Screens

The one-page guide: what goes missing without decision files, the folder structure, the habits that keep it current, where the repo sits next to Confluence, Jira, and Figma, and the prompts to try in order. Then the template every decision file follows, and how feedback travels from a review to a decision file and a change in code.

The screens are behind a password

The screens for this and the other case studies are shared with the people I am talking to. The password is in the resume I sent with my application. If you do not have it, ask me for it.